Equivariant Flow Matching for Point Cloud Assembly

TL;DR

This paper introduces an equivariant flow matching approach for point cloud assembly, enabling efficient and accurate reconstruction of 3D shapes from multiple pieces, including non-overlapping inputs.

Contribution

It presents a novel equivariant diffusion assembly model that learns related vector fields for improved point cloud assembly, with theoretical guarantees and high data efficiency.

Findings

Eda achieves high performance on practical datasets.

Eda handles non-overlapped point cloud pieces effectively.

Theoretical analysis links flow matching to learning vector fields.

Abstract

The goal of point cloud assembly is to reconstruct a complete 3D shape by aligning multiple point cloud pieces. This work presents a novel equivariant solver for assembly tasks based on flow matching models. We first theoretically show that the key to learning equivariant distributions via flow matching is to learn related vector fields. Based on this result, we propose an assembly model, called equivariant diffusion assembly (Eda), which learns related vector fields conditioned on the input pieces. We further construct an equivariant path for Eda, which guarantees high data efficiency of the training process. Our numerical results show that Eda is highly competitive on practical datasets, and it can even handle the challenging situation where the input pieces are non-overlapped.